Endogeneity in the Linear-In-Means Model

Abstract

Linear-in-means models are widely used in different contexts to estimate peer effects. In these models, there are two potential sources of endogeneity: in the interaction network and the individual’s characteristics. This paper proposes a General Three-Stage Least Square estimation modified to account for the endogeneity of the network and covariates in the linear-in-means model. The new procedure, called G3SLSX, modifies the G3SLS (Estrada et al., 2021) to recover the social and direct effects using a predetermined network and an exogenous variable as instrument. The Monte Carlo experiments show that G3SLSX has similar performance as G3SLS for the social effects. For the direct effects, G3SLSX outperforms G3SLS in the case of over-identification.